[13] RFR (S): 8225106: C2: Parse::clinit_deopt asserts when holder klass is in error state

[Vladimir Ivanov]

Thanks, Martin.
> The new assertions look correct to me. Would "initializion_not_started" be a better name?
Probably, but ciInstanceKlass::is_not_initialized() mirrors existing 
InstanceKlass::is_not_initialized(). So either both should be renamed or 
left as is.

> I'd rather use Unimplemented() in MacroAssembler::clinit_barrier when both labels are provided, because this case is not used and doesn't look very useful. But you can keep what you have if you prefer that.

Agree. I'll switch to Unimplemented() before pushing.

> I think the "assert(VM_Version::supports_fast_class_init_checks(), ..." are pointless in platform files. Would you like to keep them?

Yes, I'd like to leave them as is.

In x86_64.ad it doesn't look completely redundant:
    if (C->clinit_barrier_on_entry()) {
      assert(VM_Version::supports_fast_class_init_checks(), "sanity");

Moreover, c1_LIRAssembler_x86.cpp is shared between 32-bit & 64-bit 
ports and 32-bit port lacks fast clinit checks.

> Thanks for improving the test. Seems like you currently expect a wrong exception:
> Execution failed: `main' threw exception: java.lang.AssertionError: INIT_FAILURE: unexpected exception thrown: expected java.lang.NoClassDefFoundError, caught java.lang.AssertionError

More like the error message is a bit misleading: the failure is caused 
by a  bug which is addressed by JDK-8225141 [1]. Would providing 
exception message improve the situation?

> Also thank you for looking at my PPC64 and s390 implementations. I'll rebase and update them after this change is pushed.

Sounds good.

Best regards,
Vladimir Ivanov

[1] 
https://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2019-May/034062.html

>> -----Original Message-----
>> From: Vladimir Ivanov <vladimir.x.ivanov at oracle.com>
>> Sent: Freitag, 31. Mai 2019 22:15
>> To: hotspot compiler <hotspot-compiler-dev at openjdk.java.net>
>> Cc: Doerr, Martin <martin.doerr at sap.com>
>> Subject: [13] RFR (S): 8225106: C2: Parse::clinit_deopt asserts when holder
>> klass is in error state
>>
>> http://cr.openjdk.java.net/~vlivanov/8225106/webrev.00/
>> https://bugs.openjdk.java.net/browse/JDK-8225106
>>
>> Some asserts introduced as part of JDK-8223213 [1] don't take into
>> account the case when the class which participates in class
>> intialization barrier is in error state (class initialization failed
>> with an exception).
>>
>> Proposed fix adjusts the asserts, plus slightly improves handling of
>> cases when classes in error state are encountered.
>>
>> Testing: hs-precheckin-comp, tier1-4
>>
>> Best regards,
>> Vladimir Ivanov
>>
>> [1] https://bugs.openjdk.java.net/browse/JDK-8223213